Convolutional Neural Networks on the HEALPix sphere: a pixel-based algorithm and its application to CMB data analysis

Abstract

We describe a novel method for the application of Convolutional Neural Networks (CNNs) to fields defined on the sphere, using the HEALPix tessellation scheme. Specifically, We have developed a pixel-based approach to implement convolutional layers on the spherical surface, similarly to what is commonly done for CNNs in Euclidian space. The algorithm is fully integrable with existing libraries for NNs (e.g., PyTorch or TensorFlow). We present two applications: (i) recognition of handwritten digits projected on the sphere; (ii) estimation of cosmological parameter from Cosmic Microwave Background (CMB) simulated maps. We have built a simple NN architecture, consisting in four convolutional+pooling layers, and have used it for all the applications explored herein. For what concerns the handwritten digits, our CNN reaches an accuracy of about 95%, comparable with other existing spherical CNNs. For CMB applications, we have tested the CNN on the estimation of a "mock" parameter, defining the angular scale at which the power spectrum of a Gaussian field projected on the sphere peaks. We have estimated this parameter directly from maps, in several cases: temperature and polarization, presence of noise and partial sky coverage. In all the cases, the NN performances are comparable with those from standard spectrum-based bayesian methods. We demonstrate, for the first time, the capability of CNNs to extract information from polarization fields and to distinguish between E and B-modes. Lastly, we have applied our CNN to the estimation of the Thomson scattering optical depth at reionization (tau) from simulated CMB maps. Even without any specific optimization of the NN architecture, we reach an accuracy comparable with standard bayesian methods. This work represents a first step towards the exploitation of NNs in CMB parameter estimation and demonstrates the feasibility of our approach.

Accepted for publication on A&A

References in corpus (6)

• Measuring the Baryon Acoustic Oscillation scale using the SDSS and 2dFGRS
• Rotation-invariant convolutional neural networks for galaxy morphology prediction
• DeepSphere: Efficient spherical Convolutional Neural Network with HEALPix sampling for cosmological applications
• Polarized CMB power spectrum estimation using the pure pseudo-cross-spectrum approach
• Fast cosmological parameter estimation using neural networks
• DeepCMB: Lensing Reconstruction of the Cosmic Microwave Background with Deep Neural Networks

Cited by in corpus (28)

• DeepSphere: Efficient spherical Convolutional Neural Network with HEALPix sampling for cosmological applications
• Parameters Estimation for the Cosmic Microwave Background with Bayesian Neural Networks
• Full-sky Cosmic Microwave Background Foreground Cleaning Using Machine Learning
• OSLO: On-the-Sphere Learning for Omnidirectional images and its application to 360-degree image compression
• ForSE: a GAN based algorithm for extending CMB foreground models to sub-degree angular scales
• The Power of Halometry
• Inpainting Galactic Foreground Intensity and Polarization maps using Convolutional Neural Network
• CENN: A fully convolutional neural network for CMB recovery in realistic microwave sky simulations
• DeepSZ: Identification of Sunyaev-Zel'dovich Galaxy Clusters using Deep Learning
• Foreground model recognition through Neural Networks for CMB B-mode observations
• A Generative Model of Galactic Dust Emission Using Variational Inference
• Machine Learning and Cosmology
• Approximate Bayesian Neural Doppler Imaging
• Identifying nearby sources of ultra-high-energy cosmic rays with deep learning
• Inference of the optical depth to reionization $\tau$ from $\text{Planck}$ CMB maps with convolutional neural networks
• Extracting the gamma-ray source-count distribution below the Fermi-LAT detection limit with deep learning
• Planet cartography with neural learned regularization
• Classifying CMB time-ordered data through deep neural networks
• ForSE+: Simulating non-Gaussian CMB foregrounds at 3 arcminutes in a stochastic way based on a generative adversarial network
• A convolutional-neural-network estimator of CMB constraints on dark matter energy injection
• Deep Needlet: A CNN based full sky component separation method in Needlet space
• Self-supervised component separation for the extragalactic submillimeter sky
• Reducing false positives in strong lens detection through effective augmentation and ensemble learning
• CMB delensing with deep learning
• Exoplanet Cartography using Convolutional Neural Networks
• Constraining the polarisation flux density and angle of point sources by training a convolutional neural network
• Towards detecting Primordial non-Gaussianity in the CMB using Spherical Convolutional Neural Networks
• A novel Cosmic Filament catalogue from SDSS data